Demonstrates efficient image reconstruction in single-pixel imaging using a deep learning approach on edge AI hardware, suggesting improved real-time applications.
Key Points
The aim is to develop a real-time single-pixel imaging system using deep learning denoising on embedded GPU hardware.
Utilized compressed sensing with Hadamard patterns for image reconstruction.
Trained a compact U-Net model using mean squared error on simulated grayscale faces.
Deployed the model on a Jetson Orin NX 16 GB for efficient processing.
Evaluated performance based on acquisition, reconstruction, and inference times.